Melanoma Disease Diagnosis and Prediction Using Deep Learning Technique

Abstract

Melanoma is an aggressive skin malignancy for which early and consistent recognition is essential. This study presents a convolutional neural network (CNN)-based framework for binary classification of dermoscopic skin-lesion images and clarifies the relationship between the diagnostic model and its deployment architecture. A working set of 25,331 images from the International Skin Imaging Collaboration (ISIC) archive was organized into 18,500 benign and 6,831 malignant cases and divided by stratified sampling into 70% training, 10% validation, and 20% testing partitions. Images were resized to 224 × 224 pixels, normalized, and augmented only within the training partition. The CNN was optimized using Adam and binary cross-entropy, while K-nearest neighbours, support vector machine, Random Forest, and Gradient Boosting models were evaluated as conventional benchmarks using the same data split. The proposed CNN achieved 93.8% test accuracy, 0.94 precision, 0.93 sensitivity, 0.94 specificity, 0.93 F1-score, and an AUROC of 0.97. The Apache Kafka–Apache Spark component is used only as a deployment layer for continuous image ingestion and model inference; it does not alter the offline classification results. These findings indicate that the proposed model can support automated melanoma screening, although external multicentre validation and prospective clinical evaluation are required before routine use.